PreprintPDF Available

Abstract and Figures

The advancement of various research sectors such as Internet of Things (IoT), Machine Learning, Data Mining, Big Data, and Communication Technology has shed some light in transforming an urban city integrating the aforementioned techniques to a commonly known term - Smart City. With the emergence of smart city, plethora of data sources have been made available for wide variety of applications. The common technique for handling multiple data sources is data fusion, where it improves data output quality or extracts knowledge from the raw data. In order to cater evergrowing highly complicated applications, studies in smart city have to utilize data from various sources and evaluate their performance based on multiple aspects. To this end, we introduce a multi-perspectives classification of the data fusion to evaluate the smart city applications. Moreover, we applied the proposed multi-perspectives classification to evaluate selected applications in each domain of the smart city. We conclude the paper by discussing potential future direction and challenges of data fusion integration.
Content may be subject to copyright.
Information Fusion 52 (2019) 357–374
Contents lists available at ScienceDirect
Information Fusion
journal homepage: www.elsevier.com/locate/inffus
Full Length Article
A survey of data fusion in smart city applications
Billy Pik Lik Lau
a ,
, Sumudu Hasala Marakkalage
a
, Yuren Zhou
a
, Naveed Ul Hassan
a , b
,
Chau Yuen
a
, Meng Zhang
c
, U-Xuan Tan
a
a
Singapore University of Technology and Design (SUTD), 8 Somapah Rd, 487372, Singapore
b
Lahore University of Management Sciences (LUMS), Lahore 54792, Pakistan
c
Southeast University, Nanjing 210096, China
Keywords:
Data fusion
Sensor fusion
Smart city
Big data
Internet of things
Multi-perspectives classication
The advancement of various research sectors such as Internet of Things (IoT), Machine Learning, Data Mining,
Big Data, and Communication Technology has shed some light in transforming an urban city integrating the
aforementioned techniques to a commonly known term - Smart City. With the emergence of smart city, plethora
of data sources have been made available for wide variety of applications. The common technique for handling
multiple data sources is data fusion, where it improves data output quality or extracts knowledge from the raw
data. In order to cater evergrowing highly complicated applications, studies in smart city have to utilize data
from various sources and evaluate their performance based on multiple aspects. To this end, we introduce a
multi-perspectives classication of the data fusion to evaluate the smart city applications. Moreover, we applied
the proposed multi-perspectives classication to evaluate selected applications in each domain of the smart city.
We conclude the paper by discussing potential future direction and challenges of data fusion integration.
1. Introduction
According to UN estimates [1] , 68% of the world population would
be living in cities by 2050. Hence, managing the existing resources and
infrastructure to cater sustainable urban living conditions for the grow-
ing needs of the urban population has become ever more challenging.
Fortunately, the advancement in Information and Communication Tech-
nologies (ICT), Internet of Things (IoT), Big Data, Data Mining, and Data
Fusion is gradually paving path for the emergence of smart cities [2–4] .
In this paper, we adopt the following denition of smart city [5] :
“A city combining ICT and Web 2.0 technology with other organizational,
design and planning efforts to de-materialize and speed up bureaucratic
processes and help to identify new, innovative solutions to city manage-
ment complexity, in order to improve sustainability and livability
The integration of aforementioned technologies into various urban
domains enables city managers to equip with the necessary information
for better planning and resource management. Several cities around the
world have already been leveraging these technologies to improve the
comfort, security, mobility, health, and well-being of their citizens. To
better evaluate rapid progress and to recognize the eorts of urban plan-
ners, smart city ranking systems have been established. For instance,
IESE cities in motion index [6] has suggested 83 indicators to rank 165
Corresponding author.
E-mail addresses: billy_lau@mymail.sutd.edu.sg (B.P.L. Lau), marakkalage@mymail.sutd.edu.sg (S.H. Marakkalage), yuren_zhou@mymail.sutd.edu.sg (Y. Zhou),
naveed.hassan@lums.edu.pk (N.U. Hassan), yuenchau@sutd.edu.sg (C. Yuen), zmeng@seu.edu.cn (M. Zhang), uxuan_tan@sutd.edu.sg (U.-X. Tan).
cities over 80 countries. New York, London, and Paris are the top three
smart cities in 2018. Smart city projects in New York [7] aim to con-
sistently improve the quality of residents’ life, reduce the environmen-
tal impacts, increase the street light eciency, and enhance the water
quality. Meanwhile, the focus of Smart London Projects [8] is to collect
city wide data to provide world class connectivity, security, and smarter
streets to its residents. Digital transformation, sustainability, and urban-
ization for improving citizen services are at the cores of Paris Smart City
Projects [9] . The following up of the top smart city list includes Singa-
pore and Tokyo, which are some other notable smart cities in the world.
In Singapore, Smart Nation Project [10] has been proposed, which in-
cludes e-payment systems, smart nation sensor platform, smart urban
mobility, and smart community initiatives, with the aim to enhance
the national digital identity of its citizens. On the other hand, Tokyo
[11] aims to become the greenest city in Asia Pacic by improving the
transportation and other sectors of their economy. Local governments
in several Chinese cities [12] , such as, Shenzhen, Shanghai, Hangzhou,
and Beijing are also shaping up their cities to facilitate economic and
social development to build high income smart cities. In addition, there
are several research institutes and laboratories focusing on developing
smart city applications, which are currently leading the worldwide ef-
fort in smart domains. These include MIT Senseable Lab [13] , Future
Cities Laboratory [14] , SINTEF Smart Cities [15] , SMART [16] , etc.
https://doi.org/10.1016/j.inus.2019.05.004
Received 12 January 2019; Received in revised form 3 May 2019; Accepted 13 May 2019
Available online 13 May 2019
1566-2535/© 2019 Elsevier B.V. All rights reserved.
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
Table 1
Literature review for data fusion on smart city.
Surveys Objectives and topics covered
Khaledgi et al. [17] Provides insights on the dierent types of data
fusion techniques by exploring their concept,
benets, and challenges.
Castanedo [18] Provides an overall view on the dierent data fusion
techniques and methods. The author also reviewed
common algorithms such as data association, state
estimation, and decision fusion.
Alam et al. [19] Provides a comprehensive survey on the
mathematical model used in data fusion for specic
IoT environments.
Wang et al. [20] Proposes an IoT architecture concept to survey on
the dierent sensor data fusion techniques and also
provides an overall view on their evaluation
framework.
Zheng [21] Discusses
about dierences on fusing sources and
varying techniques for cross domain data fusion.
El et al. [22] Provides a survey on the intelligent transportation
systems, which use data fusion techniques.
Esmaeilian, B. et al. [23] Provides a throughout study on waste management
for smart city aspects with three categories: (1)
infrastructure for the collection of product lifecycle
data, (2) new adapting business model, and (3)
waste upstream separation techniques.
Da Xu et al. [24] Provides an overall view on the current state of the
industries for IoT and discusses key enabling
technologies such as communication platforms,
sensing technologies, and services.
Chen
et al. [25] Reviews the building occupancy estimation and
detection techniques while providing a comparison
between dierent sensor types for cost, detection
and estimation accuracy, and privacy issues.
Qin and Gu [26] Introduces the data fusion algorithms in IoT domains
and data acquisition characteristics.
Nowadays, communication technology is the backbone for the smart
city applications as it provides a channel for applications to trans-
fer data eortlessly. The ongoing quest for novel, more ecient, low-
latency, and cost-eective communication technologies and networks,
such as, 5G [27–29] , wireless sensor networks (WSN) [30–32] , Low
Power Wide Area Network (LPWAN) [33,34] , and Narrow Band IoT (NB-
IoT) [35,36] and their integration in smart city projects is also relentless.
These advancement has made many data sources available due to
the potential of sensors collecting data with better coverage and power
eciency of the communication platform. With the large amounts of
data becoming readily available in a smart city, data mining techniques
[37,38] are commonly used in the collected data. It helps in identifying
the essential and important data sources in the smart city applications
such as monitoring, control, resource management, anomaly detection,
etc. With the availability of parallel data sources in various smart city
domains, data fusion techniques that combine multiple data sources, lie
at the heart of smart city platform integration. The major objectives of
data fusion are to address problematic data while enhancing the data
reliability and extracting knowledge from multiple data sources. The
existing survey papers related to smart city applications or data fusion
classication are summarized in Table 1 . Majority of these review pa-
pers [18,39–41] strictly focus on one particular smart city domain or one
genre of classication perspective. In [19] , Alam et al. have conducted
a review on data fusion technique based on mathematical model in IoT
environment. Alternately in [20] , Wang et al. have described the frame-
works of data fusion within the smart city application. Interested read-
ers can follow these references for additional technical details. However,
there is only a handful of limited work to provide a multi-perspectives
approach for data fusion problems in smart cities and this literature gap
further motivates our study.
Therefore, a dierent perspective to look at data fusion in smart
city domains is necessitated by the expanding scale and scope of data
sources, data collection techniques, and data processing system archi-
tectures. In order to cater evergrowing highly complicated applications,
studies in smart city have to utilize data from various sources and evalu-
ate their performance based on multiple aspects. To this end, we propose
multiple generic perspectives with the ability to cover the entire depth
and breadth of data fusion problems in smart city. These perspectives
include data fusion objectives, data fusion techniques, data input and
data output types, data sources, data fusion scales, and platform archi-
tectures for data processing. Utilizing proposed perspectives, we provide
an overall view of classication techniques found in the seven domains
of smart city applications such as: Smart Living, Smart Urban Area Man-
agement, Smart Environment, Smart Industry, Smart Economics, Smart
Human Mobility, and Smart Infrastructure. A simple illustration of seven
application domains discussed in this paper can be found in the Fig. 1 . In
each domain, we only select notable papers to demonstrate the univer-
sality and eectiveness of our multi-perspective approach on evaluating
the data fusion techniques. Please note that we do not provide a com-
prehensive review of all the smart city applications. Afterwards, we talk
about emerging data fusion trends in smart cities, while outlining the
best practices for deploying a smart city application. In addition, data
fusion challenges in dierent smart city applications are also identied
and discussed.
To summarize, our novel contributions in this paper are three-fold
as shown below:
We propose a multi-perspectives classication to evaluate common
data fusion techniques in smart city applications.
We provide an overview of smart city application domains and dis-
cuss the common trend of data fusion techniques in each domain
utilizing proposed multi-perspectives classication.
We list down the future challenges and the ideal scenario for deploy-
ing data fusion techniques in a smart city application.
Overall, we believe that with these contributions, the readers would
have a quick grasp on the current data fusion trends in smart city re-
search without extensively going through all the details.
The rest of the paper is organized as follows: in Section 2 , we de-
ne the data fusion classication using multi-perspectives to evaluate a
smart city application. This lays a foundation for evaluating the smart
city applications leveraging data fusion techniques. In Section 3 , dier-
ent application domains of smart city based on data fusion techniques
are evaluated using the proposed multi-perspectives classication of
data fusion. In addition, a brief overall view of the current research
trend of respective domain is presented. Subsequently in Section 4 , we
discuss the ideal data fusion scenario along with potential research di-
rections/opportunities based on speculations of smart city applications
from previous section. Lastly, we conclude our works in Section 5 .
2. Data fusion classification using multi-perspectives
In this section, we identify multiple generic perspectives with the
ability to cover the entire depth and breadth of data fusion literature in
smart city applications. We use smart city single perspective data fusion
review papers [19,26] and non-smart city data fusion classication pa-
pers [18,39–41] as references. In non-smart city literature, there are four
well-known data fusion classication techniques, which are Dasarathy’s
Classication [39] , Whyte’s Classication [40] , Fusion Architecture’s
Classication [18] , and US Joint Directories of Laboratories (JDL) data
fusion classication [41] . Dasarathy’s Classication is based on the data
input and output types between data, where Whyte’s Classication fo-
cuses on the relationship between the data. JDL focuses on classifying
the fusion process according to ve processing levels. Meanwhile, the
architecture-based classication only captures the system design level
and does not consider data relationships and types. Most of the afore-
mentioned classication of the data fusion techniques are not suitable
for evaluating the applications of a smart city.
Our proposed data fusion classication approach for smart city com-
prises of six dierent perspectives (also called categories): i) data fusion
358
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
Fig. 1. List of smart city applications domain, where data fusion is commonly applied (each domain is enclosed in the dotted pink box).
objectives ( O ), ii) data fusion techniques ( T ), iii) data input and output
types ( D ), iv) data source types ( S ), v) system scales ( L ), and vi) plat-
form architectures ( P ). Within each category, we further identify vari-
ous sub-categories (also called classes). Overall, there are 30 dierent
classes. The complete list of the adopted classication indicating all the
categories and their classes is shown in Table 2 . Short reference codes
( O, T, D, S, L, P ) for each class are also included in the table for further
use in the paper. For example, O 1 refers to the data fusion objective
category and problematic data fusion class. Similarly, S 3 refers to data
source types category and participatory class.
Note that, there could be potentially more than one perspectives
(other than data sources, fusion scales, and platform architecture) for
smart city application depending on the complexity and fusion objec-
tive itself. Below, we provide further details of all the perspectives and
classes adopted in this paper.
2.1. Data fusion objectives (O)
The data fusion techniques deployed in a smart city project is inu-
enced by the objective of applications. In this paper, we have summa-
rized the four objectives as follows:
O1: Fixing Problematic Data
‘Problematic Data’ class refers to the case when the data source is
having quality issues such as, inconsistency, imperfection, disparate-
ness, etc. Data fusion could be used as an easy approach to overcome
such problems. Examples of O 1 can be found in [27,42–44] .
O2: Improving Data Reliability
Data may suer from reliability issues when it is collected in a less
ideal (less controlled) environment with high presence of noise. In
such situation, additional data sources are required to add redun-
dancy for increasing data quality to enhance data reliability. Such
situations are identied as ‘Data Reliability’ class and [45–48] ex-
hibits such pattern. In addition, security enhancement through the
Table 2
Data fusion classications for smart city applications using multi-
perspectives.
Perspective/Category Code Classes
Data Fusion Objectives O1 Fixing Problematic Data
O2 Improving Data Reliability
O3 Extracting Higher Level Information
O4 Increasing Data Completeness
Data Fusion Techniques T1 Data Association
T2 State Estimation
T3 Decision Fusion
T4 Classication
T5 Prediction / Regression
T6 Unsupervised Machine Learning
T7 Dimension Reduction
T8 Statistical Inference and Analytics
T9 Visualization
Data Input and Output Types D1 Data in Data Out (DAI-DAO)
D2 Data In Feature Out (DAI-FEO)
D3 Feature in Feature Out (FEI-FEO)
D4 Feature in Decision Out (FEI-DEO)
D5 Decision in Decision Out (DEI-DEO)
Data Source Types S1 Physical Data Sources
S2 Cyber Data Sources
S3 Participatory Data
Sources
S4 Hybrid Data Sources
Data Fusion Scales L1 Sensor Level Fusion
L2 Building Wide Fusion
L3 Inter-Building Fusion
L4 City Wide Fusion
L5 Inter-City Fusion (or Larger)
Platform Architectures P1 Edge Computation
P2 Fog / Mist Computation
P3 Cloud Computation
P4 Hybrid Computation
359
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
data fusion also belongs to this category and examples of such ob-
jectives can be found in [49–51] .
O3: Extracting Higher Level Information
Data mining advancement has contributed to many dierent archi-
tectures of data fusion in order to obtain knowledge from multiple
data sources. For instance, the occupancy of a building can be de-
tected using a combination of few ambient sensors with data fusion,
where occupancy information cannot be directly inferred from the
raw data sources. We classify these approaches as ‘Higher Level In-
formation Extraction’ class and examples can be found in [52–54] .
O4: Increasing Data Completeness
In a situation of coverage limitations, an individual data source is
insucient to provide complete details of the desired output. There-
fore, in ‘Data Completeness’ class, data fusion is performed across
multiple data sources to obtain a complete picture of the overall sys-
tem such as [55–57] .
2.2. Data fusion techniques (T)
In this category, we present the data fusion techniques in two dier-
ent information enrichment obtained after data fusion. The T 1 until T 3
are the common data fusion techniques and the further details can be
found in [19,39] , where it describes the lower level information being
fused to generate identical level of information. The techniques 𝑇 4 𝑇 8
are associated with data mining [38,58] , where simple input data from
multiple sources is fused to generate higher level information enrich-
ment. Brief description of these classes is given below:
T1: Data Association
Data association refers to data fusion technique that fuse data based
on similarity between at least two or more data sources. Com-
mon techniques for data association include Nearest Neighbors [59] ,
Probabilistic Data Association [60] , and Multiple Hypothesis Test
[61] .
T2: State Estimation
State estimation indicates the usage of multiple data sources to
achieve higher sate estimation accuracy. Common techniques un-
der this category are Maximum Likelihood [62] , Kalman Filter [63] ,
Particle Filter [64] , and Covariance Consistency Model [65] .
T3: Decision Fusion
Decision fusion is a technique that is used to fuse the decisions made
by various sub-components of a system to achieve a certain overall
objective. For instance, a robot can fuse dierent decisions from the
modules to perform an actuation (direction, events, or actions). Gen-
eral techniques include Bayesian inference [66] , Dempster–Shafer
Inference [67] , and semantic approaches [68] .
T4: Classification
Classication technique denotes methodology of grouping objects
into dierent classes based on their unique characteristics. In-depth
details of generic classication techniques can be found in [38,58] .
T5: Prediction
Prediction techniques are used to forecast output based on single or
multiple dierent data sources. Note that, this covers simple meth-
ods such as regression and as well as complicated methods such as
forecast modeling. Examples of such can be found in [69–71]
T6: Unsupervised Machine Learning
Unsupervised machine learning tries to automate the knowledge dis-
covery without relying on the data labels. Examples of such methods
involves clustering [72] , anomaly detection [73] and others [38] .
Note that, semi-supervised machine learning approach [74] is also
categorized under this class.
T7: Dimension Reduction
Dimension reduction refers to the method of reducing data sources’
dimensions for features extraction or visualization purposes. Exam-
ples of dimension reduction techniques are Principal Component
Analysis (PCA) [75] , and others [38] . The aim is to preserve the
characteristic of the data sources while reducing the complexity of
processing high dimensional data.
T8: Statistical Inference and Analysis
Statistical inference and analysis is used for outlining certain infor-
mation along with some common knowledge / hypothesis from the
input data sources. Examples of papers using such approaches can
be found in [76,77]
T9: Visualization
Visualization is a technique used for the presentation of output to
the end users via some platform. The end result often requires hu-
man intervention. Examples of such techniques can be referred to
the following papers [78–80] .
2.3. Data input and output types (D)
Dasarathy’s classication [39] is based on input and output of fu-
sion technique to determine the relation between input and output data.
There are ve classes in data input and output perspective. Brief details
are given below:
D1: Data In Data Out (DAI-DAO)
Data In Data Out (DAI-DAO) refers to the situation when multiple
raw data sources are fused to increase data reliability and the output
after fusion is still a raw data.
D2: Data In Feature Out (DAI-FEO)
Data In Feature Out (DAI-FEO) refers to the situation when multiple
raw data sources are fused to extract some unique feature of the
observed system. The output feature describes certain aspect of the
system and it could be further used for more feature extraction or to
make certain decisions.
D3: Feature In Feature Out (FEI-FEO)
Feature In Feature Out (FEI-FEO) refers to the situation when multi-
ple unique features from dierent sensors are combined to generate
new features. This class is commonly known as feature fusion.
D4: Feature In Decision Out (FEI-DEO)
Feature In Decision Out (FEI-DEO) refers to the situation when cer-
tain features of the system are fused to make certain decisions, e.g.
actuation of various system components.
D5: Decision In Decision Out (DEI-DEO)
Decision In Decision Out (DEI-DEO) refers to the situation when dif-
ferent decision sources (maintenance status, events, etc.) are com-
bined to obtain a nal output decision.
2.4. Data source types (S)
There are four types of generic data sources in smart city applica-
tions and we categorize them based on the data sources regardless of
the communication medium. Details of each category can be found as
follows:
S1: Physical Data Sources
The physical data sources are collected from sensors that are be-
ing deployed to capture information of a particular space, area, or
even city wide. Examples of the physical sensors include temper-
ature [81] , air quality [82] , camera [83] , ultrasonic [84] , LiDAR
[85] , and etc. Note that, we categorize smart city application based
on the data sources rather than the method they are acquired. For
instance, a temperature probe in a sensor nodes of a wireless sensor
network (WSN) transmits data through gateway to cloud database is
considered as physical data source, S 1.
S2: Cyber Data Sources
Cyber data sources denote datasets which are commonly obtained
from the Internet domain such as social media information [76,86] ,
web access data [87,88] , and opinion based datasets [89] . Social me-
dia information involves major social media platforms such as Twit-
ter, Facebook, LinkedIn, Weibo, and others. Note that, usually the
data is acquired through data mining techniques. Meanwhile, the
360
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
web access data can be obtained from web applications program-
ming interface (API), such as transportation tickets information and
online customer records. Apart from that, open datasets refer to data
from third party vendors such as telecom operator or a company with
readily available data.
S3: Participatory Data Sources
Participatory data sources include crowdsensing [90,91] and crowd-
sourcing [92,93] data contributed by the personal devices, e.g. mo-
bile phones, wearable devices, tablets, etc. of the users in smart city.
Users provide the data voluntarily or through some incentive mech-
anisms.
S4: Hybrid Data Sources
The hybrid data sources include data obtained from mixed data
sources [94,95] , e.g. by combining the participatory and physical
sensor data. As pointed in [21] , hybrid data sources can achieve
more insights as compared to single data sources.
2.5. Data fusion scales (L)
The scale of data fusion is also an important classication perspec-
tive. Please note that data fusion scale is based on sensor coverage rather
than sensor deployment. There are four dierent classes, which are de-
scribed below:
L1: Sensor Level Fusion
At the sensor scale, data from various physical sensors is fused to
form an output such as [53,96] . For instance, fusion of data collected
by various smartphone sensors is an example of data fusion at sensor
level.
L2: Building Wide Fusion
At the building wide scale, data sources collected within a premise or
building is fused to form an output. For instance, fusion of building
energy and building security data to develop a building management
system [97–99] is an example of data fusion at building level.
L3: Inter-Building Fusion
In the inter-building scale, the data sources collected over several
buildings are fused to form an output, where the scale of deployment
normally includes small area. For example, data sources of several
buildings within a university are used to generate a particular out-
put is considered as inter-building scale. Other examples of this data
fusion scale also can be found in [100,101] .
L4: City Wide Fusion
In the case of city wide fusion, data sources that involve whole city’s
area as input for the data fusion architecture fall under this class such
as [102–104] . For instance, the study of citizen behavior involves
fusion of data gathered in dierent areas of the city is considered
city wide data.
L5: Inter-City Fusion (or larger)
At the inter-city fusion (or larger) scale, data from large areas in-
volving one or more cities or terrains (mountains, sea, forests, etc.)
is fused to form an output. Examples of this scale involve comparing
one smart city to another city or data of a city outskirts and its sur-
rounding areas. More examples of inter-city fusion (or larger) can be
referred to [43,105,106] .
2.6. Platform architectures (P)
The architecture of computational platform involved in data fusion is
another important classication perspective. In this category, we iden-
tify four generic classes:
P1: Edge Computation Platform
In edge computation platform, data sources are processed and fused
at the edge (i.e. very close to the physical location, where data
is actually collected). Edge computation devices include micro-
controller, computing devices (Raspberry pi), computers, etc. Such
architecture can be found in works such as [96,99,101] . With this
architecture, communication overheads and latency can be signi-
cantly reduced.
P2: Fog Computation Platform
In fog computation platform, data sources are processed and fused
at the middle layer, i.e. between the edge and the cloud. In this ar-
chitecture, data is periodically or continuously sampled at the edge
(without processing) and is then forwarded to a gateway (that acts
as a fog device). At the gateway, computing resources are provided
for data processing. Both fog computing and edge computing plat-
forms provide similar benets of ooading computation as shown in
[102,107,108] . However, fog computing architecture should be pre-
ferred when it is dicult to nd stable power sources at the edge.
P3: Cloud Computation Platform
In cloud computation platform, data sources are processed and fused
in the cloud. This is the most common technique practiced by indus-
try and research institutes for processing big data. Examples of this
architecture being used are [56,87,109] . The advantages of cloud
computing architecture includes ready access to the data and both
online and oine for further processing or fusing. The disadvantages
include increased communication overheads and costs.
P4: Hybrid Computation Platform
In hybrid computation platform, processing is distributed among two
or more layers (edge, fog and cloud) as shown in [105,110,111] . In
this architecture, depending on the available resources or applica-
tion objectives, some low level data fusion and processing is done
at the edge or fog, while high level information is extracted in the
cloud.
3. Smart city applications overview
Smart city applications tend to have extremely diverse requirements,
which contribute to a large variety of dierent techniques and require-
ments as stated previously in Section 2 for dierent domains. Thus, it
is necessary to evaluate the smart city applications from a more generic
perspectives rather than one specic perspective. In this section, we se-
lect smart city applications with data fusion techniques from dierent
domains listed in Fig. 1 , and evaluate them based on multi-perspectives
from the Section 2 . Note that, there exist some literatures that are
cross-disciplinary, which may involve more than one domain. In or-
der to address the cross-disciplinary smart city applications, we have
grouped them into their closest relevant domain. In each application
domain, we outline sub-domains and present works related to data fu-
sion techniques. Using the proposed data fusion classication based on
multi-perspectives, we discuss the common data sources and fusion tech-
niques, along with the current research trends in each domain.
3.1. Smart living
Smart living concerns with the life of the urban citizens and revolves
around the concept of improving live-ability in urban area. In the litera-
ture, the general objectives of utilizing the smart living domain involve
data being used to extract higher level information or increasing the data
completeness. In addition, smart city applications in this domain often
leverage the cloud or hybrid platform architecture. In this domains, we
have studied three dierent aspects of smart living, namely, (1) Smart
Health, (2) Smart Home, and (3) Smart Community ( Table 3 ).
3.1.1. Smart health
Healthcare is a crucial component in everyday life concerning medi-
cal and public practices using devices as dened by Lee and Co-authors
[144,145] . The rapid development of technology (e.g. smartphones and
their in-built sensing devices such as heart rate sensors) provides more
opportunities to adopt technology in healthcare applications perva-
sively. For telehealth application in smart city, Hossain et al. [112] have
used electroencephalographic (EGG) signals and voice to monitor a spe-
cic user’s health with the support of cloud technology and doctor’s
361
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
Table 3
List of smart city applications using data fusion technique(s).
Domain Sources O S D T L P Remarks
Smart Living [52] 3 1 2 4 1 4 Smart Healthcare
[112] 3 1 1 4 4 4 Voice Pathology Detection
[113] 3 3 4 3 2 4 Smart Home Healthcare Monitoring
[110] 3 1 2 4 1 4 Daily Activity Classication
[45] 2 1 2 4 2 4 Smart Home Activity Recognition
[111] 3 1 3 4 2 4 Tele-Rehabilitation
[114] 4 4 4 3 2 4 Smart Home Control System
[79] 3 4 3 9 4 4 Intelligent Video Surveillance
[115] 3 3,4 4,5 4,5 5 3 Distance Learning
[116] 3 2 3 1 4 3 Smart Community
Smart
Urban Area Management [94] 4 4 4 5 2 3 Building Management
[117] 4 1 2 2 1 1 Fire Detection System
[118] 3 3 4 3 5 3 Lean Government
[43] 1 1 1 1 5 1 Urban Planning with Satellite Images
[95,119] 3 4 1,2 8,9 4 3 Urban Space Utilization Detection
[56] 4 3 1 9 4 3 Fault Reporting Platform
[76] 3 4 3 8,9 5 3 Landscape Rating Systems
Smart Environment [106] 3 1 1 9 5 3 City Environment Monitoring
[78] 3 1 2 2,9 1 1 City Building Map Modeling
[120] 3 4 4 4 5 1 Forest Types Classication
[46] 2 1 1 1 5 1 Long Term Landscape Monitoring
[121] 4 1 4 4 5 1 Forest Species Classication
[122] 3 4 2 4 1 1 Waste Water Treatment
[102] 4 1 2,3 2,9 4 2 Urban Solid Waste Management
Smart Industry [96,123] 4 1 2,4 2,4 1 1 Fault Detection
[124,125] 3,4 1 3,4 5 1 1 Tools Life Prediction
[98] 2 4 4,5 3 2 1 Decision Support in Manufacturing
[51] 2 1 2,4 2,3 1 1 Autonomous Robots and Security
[126] 3 1 2,4 4,7 1 1 Seafood Freshness Classication
[127,128] 3,4 1 2,4 2,4,5 1 1 Agriculture Plant Disease Classication
Smart Economics [87] 4 2,3 1,3 1,8 5 3 Customer Proling
[129] 4 4 1,4 5 5 3 Consumer Awareness
[107] 4 4 1 9 5 2 Blockchain and Supply Chain
[130] 3,4 4 2,4 1,5,8 5 3 Supply Chain Management
[77] 3 2,3 2,3 5,8 4 3 Tourist Behavior Analysis
[57] 4 4 2,3 6 4 3 Travel Recommendation System
[131] 3 1,2,3 2 1,4 4 1,3 Tourist Tracking Application
Smart Human Mobility [132,133] 2,3 1 1 5,1 4 3 Outdoor Positioning
[134,135] 2,4 1 1 1,2 2 3 Indoor Positioning
[136,137] 4 1 4 5,1 4,2 3 Location-based Services
[103] 3 3 2 1 4 3 Obtaining Origin-Destination Matrices
[54] 3 3 2 4 4 3 Identifying Transportation Modes
[138] 3 3 2 1 2 3 Monitoring Visitors Inside a Building
[109] 4 1 4 3 4 3 Trac Signal Controlling
[139] 3 3 2 1 4 3 Analyzing Public Transport Services
[140] 4 1 4 4 1 3 Autonomous Vehicle Controlling
Smart Infrastructure [55,88] 3,4 1 2,4 5 4,1 1 Smart Grid and Power Utilities
[101,141] 3 1,4 1 4,5 3 1 Solar Farm
[105] 3 3 2 2 5 4 Smart Metering
[27,42] 1,2 1 1 1,2 1 1,3 Communication (5G)
[47,48] 2 1 1 1,5 1 1 Communication (WSN)
[142] 4 1 2,3 4 1 1 Drone Detection
[143] 3 4 1,2 2 4 3 Smart Parking System
[99] 4 1 1 2 2 1 Bridge Monitoring Platform
[104] 3 1 2,3 4,5 4 3 Water Distribution System
advices. In
[113] , work has shown to monitor elderly at home based on
fuzzy fusion model using behavioral and acoustical environment data.
Similarly, Noury [146] also monitors the activities and fall detection
of elderly through fuzzy logic by fusing accelerometer, vibration, and
orientation sensor. In [91] , Marakkalage et al. have used crowd-sensing
data from a smartphone application (location, noise, light, etc.) and in-
troduced sensor fusion based environment classication (SFEC) to pro-
le elderly people for understanding their daily lifestyle. In addition,
Dawar and Kehtarnavaz [52] have implemented a Convolution Neural
Network (CNN) to combine both depth camera and wearable devices to
detect the transition of movements to fall. Apart from that, Hondori et al.
[111] have proposed using sensor fusion between depth images and in-
ertia to perform tele-rehab in the home. The main challenge occurs in
pervasive smart healthcare data fusion is discussed in [147] as the need
of a higher accuracy to improve sensing robustness against uncertainty
and unreliable integration.
3.1.2. Smart home
The concept of Smart Homes plays an important role nowadays in
contemporary urban areas. According to Jiang et al. [148] , the deni-
tion of a smart home provides the capability of controlling, monitor-
ing, and accessed appliances & services through implementation of ICT.
There are currently many big players in developing the smart home ap-
pliances such as Amazon, Google, Apple, IBM, Intel, Microsoft, Xiaomi,
and others. The challenge faced by manufacturers are related with ser-
vice integration and formulating software ontology platform. These are
necessary for implementing the services through dierent vendors and
allow for a better integration. Meanwhile in [114] , physical sensors (soil
moisture) and cyber (weather, trac) have been fused to control home
appliances such as alarm clock and water sprinkle. The study of user
daily activity is yet another important aspect to understand urban citi-
zen well-being. In [45] , Hong et al. have combined series of life activities
to understand the lifestyle pattern depends on the equally weighted sum
362
B.P.L. Lau, S.H. Marakkalage and Y. Zhou et al. Information Fusion 52 (2019) 357–374
operation and Dempster-Shafer theory. Also, similar study on the user
daily activity patterns can be found in [110] . Combination of house
environmental sensor (infrared, door contact, temperature, hygrome-
try sensor, microphone) and wearable devices (kinematic sensors) using
support vector machine (SVM) can be used to identify the user activity
patterns. In addition, the modeling of human behavior in a smart home
[149] in order to generate learning situation models have proven the
eciency of context-aware services. In addition, smart home security is
yet another study eld for many researchers [150–152] due to increased
usage of IoT devices in normal household. The research challenges is to
develop the applications for the smart houses while retaining the pri-
vacy and security of the end user.
3.1.3. Smart community
According to Smart Communities Guidebook [153] , a smart commu-
nity is described as “a geographical area ranging in size from neighbor-
hood to a multi-county region whose residents, organizations, and gov-
erning institutions are using information technology to transform their
region in signicant ways ”. There is only a handful of cities focus on
this aspect as majority are still in the stage of transforming from facil-
ity to community welfare. First world countries such as USA, Canada,
Australia, European Union, and Singapore shown in [154] have started
up initiatives to create smart communities. Information fusion for smart
community video surveillance system is performed in [79] to aid neigh-
borhood in terms of security. The combination of the dierent modal
surveillance camera provides a vast amount of visual information extrac-
tion such as video summarization for highlighting certain events. A dis-
tance learning framework is proposed in [115] , which enables personal-
ized learning to cater what is best for each individual user. It uses data
fusion to understand user environment and their activities by means
of hybrid data sources. Real-time community monitoring also helps to
prevent emergency situations and it ensures the safety of community cit-
izens. A good example for a smart community application in large-scale
is the Social Credit System in China [155] . It is a state-owned system
to collect data from both public (trac cameras, transit data etc.) and
private (online shopping, tness trackers etc.) data sources to monitor
and analyze user behaviour to generate a single ”credit score ”for each
person, which helps in community well-being. The techniques fuse these
data sources and remains a back box to the general public. However, the
eect on user privacy with the rise of “data state ”remains a debate for
some [156] . A mature citizen should be on alert and always responds
to any potential threat, while spreading the awareness to build a safer
community in the urban city.
3.2. Smart urban area management
Smart urban area management denotes the managing of urban area
using ICT. Sub-domains in this regime composed of urban planning, gov-
ernance, and smart buildings. For an application to t into this deni-
tion, the minimum scale would be at the building level (e.g. a building
management system). The main trend of data fusion techniques being
applied in this domain mostly consists of objectives of extracting higher
level information or increasing the data completeness. The end product
of data fusion include visualization of information for respective author-
ities.
3.2.1. Smart governance
In smart governance, managing a city is considered as a complex
task as the integration of dierent domains and services is proven to
be challenging. Transparent services integration is an example of why
many governance authorities are having diculties to sort it out. It is
hard to strike a balance in developing a transparent governance policy
with consideration of sensitive information. Therefore, there is only lim-
ited study materials available to the best of our knowledge. Janssen and
Estevez [118] have proposed a centralized platform for cutting down
government sta by shifting existing organization to rely on integration
of platforms. The disaster response management is also considered as
another vital element for a smart city to carry out any potential counter
measurements towards disaster as shown in [157] . Apart from that, ur-
ban reporting system [56] has collected report from the city wide region
on the faulty infrastructure so that immediate actions can be taken to
remedy the situation. It uses cloud technology and focuses on the dis-
play of fused data report, which it also describes the location and types
of infrastructure. Example of research challenges is to remove any po-
tential fake report to prevent misuse of the reporting platform. Another
example of smart governance that involves city safety can be found in
[158] , where it can act as an emergency aid application (light pulse
on emergency through mesh network) while providing energy ecient
lighting to urban area. Moreover, there are cities also working on gover-
nance platform such as New York [7] , Singapore [10] , Tokyo [11] , Oslo
[159] , and others. The potential research opportunity is to propose con-
sensus protocols within the city for better integration of services.
3.2.2. Smart urban planning
Urban planning plays an important role in developing the city econ-
omy by taking account of well-being of the urban residents. Tradition-
ally in urban planning, aerial photography and statistical data sources
(building size, population number, public amenities, etc.) are combined
to understand the current development state of the city. The downside
of such method is data sources frequently lacks of ne details, which
resulting the output result is not representative. To address such issue,
Cheng and Toutin [43] have combined various satellite and aerial im-
ages to generate details for the exiting urban structures. Alternately, low
power sensors are capable to provide a larger coverage with lower de-
ployment cost, which give researchers the opportunity to study dierent
points of interest in the urban area. In [81,95,119] , a bottom up urban
planning method is implemented, where sensors are installed in a desig-
nated region to capture space utilization. From the collected data, urban
planners can study public space utilization pattern using an integrated
portal. Here, a hybrid processing method is proposed, where the data
processing and fusion occur in dierent stages of data pipeline. In addi-
tion, a large variety of data sources can be used for urban planning such
as physical sensors [160] , photography [76,161] , or hybrid data sources
[85] . Despite wide variety of data sources, human interpretation is re-
quired when it comes to make decision on a proposed urban design. The
need of full automated planning system would further benet the urban
planners to combine dierent data sources in order to achieve a more
ideal city planning.
3.2.3. Smart building
Urban building management provides building owner a platform to
understand building’s energy consumption rate while automating build-
ing resources management. It has been extensively studied in [25,162–
164] and the current trend is to optimize the building resources such as
hot water systems, electrical consumption, and heating ventilation & air
conditioning (HVAC). In [94] , Aftab et al. have combined four dierent
parameters to predict building occupancy to control HVAC using low-
cost embedded systems. Some other works such as [97,165,166] also
have the same objectives but using dierent types of data sources. The
potential solution for better building management system is to rely on
fusing weather, human feedback, and electricity price to ne tune the
building resources in order to maximize human comfort, while mini-
mizing the energy consumption. Apart from that, re alarm system is
considered another important features of the smart building manage-
ment system. Luo and Su [117] have fused three dierent data sources
(ame, smoke, and temperature sensor) to detect any potential re out-
break and reduce false alarms. In addition, a notication-based system
is implemented to notify the property owner and manager in case of
emergency. In future, potential building safety features may include a
group of robots to deal with re hazards and double duty as building
security patrols.
363